Abstract

Replacing the triangle inequality by
$\parallel x+y{\parallel}^{2}\le 2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})$
in the definition of norm we obtain the notion of parallelogram norm. We establish that every parallelogram norm is a norm in the usual sense. 1 Introduction

The parallelogram law states
$\parallel x+y{\parallel}^{2}+\parallel x-y{\parallel}^{2}=2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})$
for any vectors
$x$
and
$y$
in a Hilbert space. Then inequality
$\parallel x+y{\parallel}^{2}\le 2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})$
trivially holds. S. Saitoh [2 ]
$$
noted that as triangle inequality the inequality
$\parallel x+y{\parallel}^{2}\le 2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})$
may be more suitable. He generalized this inequality to the setting of a natural sum Hilbert space for two arbitrary Hilbert spaces.

Obviously the classical triangle inequality in an arbitrary normed space implies the inequality above. This motivates us to introduce the notion of parallelogram norm.

Let
$X$
be a real or complex linear space. A mapping
$\parallel .\parallel :X\to [0,\infty )$
is called a parallelogram norm on
$X$
if it satisfies the following conditions:

(i)
$\parallel x\parallel =0$
if and only if
$x=0$
; (ii)
$\parallel \alpha x\parallel =\left|\alpha \right|\parallel x\parallel $
; (iii)
$\parallel x+y{\parallel}^{2}\le 2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})$
for all
$x,y\in X$
and all scalar
$\alpha $
.

As mentioned before, every parallelogram norm is a norm in the natural sense. We aim to prove the converse. The reader is refered to [1 ]
$$
for undefined terms and notation.

2 The Result

We start this section with the following lemma.

Lemma 2.1
Let
$X$
be a real or complex linear space and
$\parallel .\parallel :X\to [0,\infty )$
be a map satisfying (i), (ii) in the definition of a parallelogram norm. Then
$\parallel .\parallel $
is a norm if and only if the set
$\mathbf{B}=\{x|\parallel x\parallel \le 1\}$
is convex.

Proof

If
$\parallel .\parallel $
is a norm, then B is clearly a convex set. Conversely, let B be convex and
$x,y\in X$
.

We can assume that
$x,y\ne 0$
. Putting
${x}^{\prime}=\frac{x}{\parallel x\parallel}$
and
${y}^{\prime}=\frac{y}{\parallel y\parallel}$
we have
${x}^{\prime},{y}^{\prime}\in \mathbf{B}$
. Now
$\lambda {x}^{\prime}+(1-\lambda ){y}^{\prime}\in \mathbf{B}$
for each
$0\le \lambda \le 1$
. In particular, for
$\lambda =\frac{\parallel x\parallel}{\parallel x\parallel +\parallel y\parallel}$
we obtain
$$\parallel \frac{x}{\parallel x\parallel +\parallel y\parallel}+\frac{y}{\parallel x\parallel +\parallel y\parallel}\parallel =\parallel \lambda {x}^{\prime}+(1-\lambda ){y}^{\prime}\parallel \le 1.$$
So that
$\parallel x+y\parallel \le \parallel x\parallel +\parallel y\parallel .\square $
We are just ready to prove our main result.

Theorem 2.2
Each parallelogram norm
$\parallel .\parallel $
is a norm in the usual sense.

Proof

We have to show that
$\mathbf{B}=\{x|\parallel x\parallel \le 1\}$
is convex. Let
$x,y\in \mathbf{B}$
. Then we have
$$\parallel x+y{\parallel}^{2}\le 2(\parallel x{\parallel}^{2}+\parallel y{\parallel}^{2})\le 2(1+1)=4,$$
whence
$\parallel \frac{x+y}{2}{\parallel}^{2}\le 1$
, so
$\frac{1}{2}x+(1-\frac{1}{2})y\in \mathbf{B}$
. Thus if
$A=\{\frac{k}{{2}^{n}}|n=1,2,\cdots ,k=0,1,\cdots ,{2}^{n}\}$
then for each
$\lambda \in A$
we have
$\lambda x+(1-\lambda )y\in \mathbf{B}$
.

Let
$0<\lambda <1$
and
$z=\lambda x+(1-\lambda )y$
. Since
$A$
is dense in
$(0,1)$
, there exists a decreasing sequence
$\left\{{r}_{n}\right\}$
in
$A$
such that
${lim}_{n\to \infty}{r}_{n}=\lambda $
. Put
${\beta}_{n}=\frac{1-{r}_{n}}{1-\lambda}$
. Obviously
$0<{\beta}_{n}<1,{lim}_{n\to \infty}{\beta}_{n}=1$
and
$\frac{{r}_{n}+{\beta}_{n}-1}{{r}_{n}}<1$
. Since
$\frac{{r}_{n}+{\beta}_{n}-1}{{r}_{n}}x\in \mathbf{B}$
and
${r}_{n}\in A$
we conclude that
$${\beta}_{n}z={\beta}_{n}\lambda x+(1-\lambda ){\beta}_{n}y={r}_{n}\left(\frac{{r}_{n}+{\beta}_{n}-1}{{r}_{n}}\right)x+(1-{r}_{n})y\in \mathbf{B}.$$
Thus
$\left|{\beta}_{n}\right|\parallel z\parallel =\parallel {\beta}_{n}z\parallel \le 1$
for all
$n$
. Tending
$n$
to
$\infty $
we get
$\parallel z\parallel \le 1$
, i.e.
$z\in \mathbf{B}.\square $
Acknowledgment. We would like to sincerely thank Professor Saburou Saitoh for his encouragement.

References

- W. B. Johnson (ed.) and J. Lindenstrauss (ed.), Handbook of the geometry of Banach spaces, Vol. 1, North-Holland Publishing Co., Amsterdam, 2001. x+1005 pp.
- S. Saitoh, Generalizations of the triangle inequality, J. Inequal. Pure Appl. Math. 4 (2003), No. 3, Article 62, 5 pp.